Abstract
We propose a novel method for training neural networks to predict the future prices of stock indexes. Unlike previous works, we do not use target stock index data for training neural networks for index prediction. Instead, we use only the data of individual companies to obtain sufficient amount of data for training neural networks for stock index prediction. As a result, our method can avoid various problems due to training complex machine learning models on a small amount of data. We performed numerous types of experiments to test methods designed for predicting the future price of the S&P 500 which is one of the most commonly traded stock indexes. Our experiments show that neural networks trained using our method outperform neural networks trained on stock index data. Compared with other state-of-the-art methods, our method is conceptually simpler and easier to apply, and achieves better results. We obtained approximately a 5-16% annual return before transaction costs during the test period (2006-2018).
Highlights
Predicting future prices of stock indexes such as the S&P 500 or the Nasdaq Composite is a challenging and important task
We proposed a novel method for training various types of Neural Networks (NNs) to predict the future price of the S&P 500, one of the most commonly traded stock indexes
We trained the target NNs only on the data of individual companies, which is a sufficient amount of data; this helped avoid problems due to training NNs on a small amount of data
Summary
Predicting future prices of stock indexes such as the S&P 500 or the Nasdaq Composite is a challenging and important task. We argue that only several thousands of data points are an insufficient amount of data for effectively training complex NNs to predict the future prices of stock indexes. In various fields, such as natural language processing [11] or image classification [12], many works have recently shown that training NNs on a larger amount of training data increases the performance of models. In our experiments, we empirically show that when building NNs for stock index prediction, training the NNs on the data of individual companies is more effective than training on the data of stock index. We consider transaction costs in our experiments and introduce a simple method for controlling the number of transactions
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
